chore(*): culmulative AI GW fixes 20251028#14785

Open

fffonion wants to merge 54 commits intomasterfrom

Commits on Oct 30, 2025

chore(llm): move keybastion to shared module
fffonion
committed
perf(llm): make parsing sse chunk 50% faster (#12574)

spacewander
authored and
fffonion
committed
fix(analytics): fixed an issue where some of the ai metrics not included (#12505)

liverpool8056
authored and
fffonion
committed
chore(*): fix a typo in ai-proxy plugin
fffonion
committed
fix(llm): every chunk should end with \n\n (#12468)

committed
fix(ai-bedrock): do not set api dimensionality for bedrock titan embeddings

oowl
authored and
fffonion
committed
fix(ai): set bedrock single-string tool response in correct field

oowl
authored and
fffonion
committed
fix(ai): ollama correct handling of tools and timestamp formats

oowl
authored and
fffonion
committed
fix(ai-proxy): fix AI-response-transformer plugin subrequest llm API return 403

dylankyc
authored and
fffonion
committed
fix(ai-proxy): resp for observability may be larger than the real one

spacewander
authored and
fffonion
committed
fix(ai-proxy): fix AI Plugins with multiple capture groups

dylankyc
authored and
fffonion
committed
fix(ai-proxy): Implement latency metric for streaming responses (#12837)

spacewander
authored and
fffonion
committed
fix(ai-proxy): large request payload was not logged (#12915)

spacewander
authored and
fffonion
committed
fix(ai): support anthropic models in gemini (#12789)

authored and
fffonion
committed
fix(gemini): fix non-chat request handling for gemini driver
fffonion
committed
chore(ai-proxy): drop duplicate get_source check

spacewander
authored and
fffonion
committed
chore(*): address conflicts of normalize-sse-chunk
fffonion
committed
fix(llm): fix error of failover for bedrock driver
fffonion
committed
refactor(llm): move save-request-body to a standalone filter
fffonion
committed
fix(llm): fix ai plugin retry phase kong.plugin.ctx missing correct namespace (#13224)

oowl
authored and
fffonion
committed
fix(llm): AI Proxy might produce duplicate content in the response when the SSE event was truncated (#13315)

spacewander
authored and
fffonion
committed
fix(embeddings): fix gemini vertex embeddings
fffonion
committed
fix(llm): AI Proxy might drop content in the response when the SSE event was truncated (#13452)

spacewander
authored and
fffonion
committed
fix(ai-proxy): aws stream parser didn't parse correctly when the frame was incomplete (#13430)

spacewander
authored and
fffonion
committed
fix(llm): fix ai driver set wrong target port when upstream using different http scheme (#13288)

oowl
authored and
fffonion
committed
fix(ai-proxy): JSON array is not correctly parsed for Gemini and Cohere events (#13588)

spacewander
authored and
fffonion
committed
fix(ai): gemini and anthropic correct stop_reason mappings

tysoekong
authored and
fffonion
committed
fix(tests): ai tests move to compare whole object

tysoekong
authored and
fffonion
committed
fix(ai): AG-405: add mappings for gemini structured output

tysoekong
authored and
fffonion
committed
fix(ai): AG-405 add anthropic structured output support

tysoekong
authored and
fffonion
committed
fix(ai): AG-405 add bedrock structured output support

ttyS0e
authored and
fffonion
committed
fix(ai): mistral unknown field seed error

tysoekong
authored and
fffonion
committed
tests(e2e): add tests for structured output

tysoekong
authored and
fffonion
committed
chore(llm): add error handling and tests for iso_8601_to_epoch
fffonion
committed
fix(ai-proxy): map OpenAI chat completion's tool_choice when using Anthropic provider (#13355)

spacewander
authored and
fffonion
committed
feat(gemini): add vertex ai model garden model support (#13734)

oowl
authored and
fffonion
committed
fix(llm): fix content_type judgment typo

oowl
authored and
fffonion
committed
fix(ai-proxy): `model` field was missing in OpenAI format response from Gemini provider in some situations

spacewander
authored and
fffonion
committed
fix(ai-proxy): `model` field was incorrect in Gemini responses with var used as model name

spacewander
authored and
fffonion
committed
fix(ai-proxy): empty array in structured output was encoded as empty object

spacewander
authored and
fffonion
committed
fix(ai): bedrock structured_output streaming responses

tysoekong
authored and
fffonion
committed
fix(ai): bedrock not supporting multiple toolUse in one turn

tysoekong
authored and
fffonion
committed
fix(ai): refactor gemini transformer to properly handle function calling in llm/v1/chat

tysoekong
authored and
fffonion
committed
fix(ai): anthropic streaming tool_use responses

tysoekong
authored and
fffonion
committed
fix(ai): anthropic streaming tool_use responses

tysoekong
authored and
fffonion
committed
fix(ai): truncation of anthropic streaming responses when in mid-struct

tysoekong
authored and
fffonion
committed
feat(huggingface): add support for new serverless inference provider and happy path for e2e test (#13679)

dylankyc
authored and
fffonion
committed
fix(ai-proxy): fix huggingface embedding inproperly parsed and add e2e test (#14153)

dylankyc
authored and
fffonion
committed
fix(ai): AG-488 Correctly return GCP Model Armor 'floor' failures [aigw-only] (#14137)

tysoekong
authored and
fffonion
committed
fix(ai-proxy): fix huggingface inference provider does not support use_cache and wait_for_model options [aigw-only] (#14287)

dylankyc
authored and
fffonion
committed